1. Field of the Invention
The present invention relates to amplifiers using metal oxide semiconductor based integrated circuits. The present invention is particularly but not exclusively related to audio application mixed signal chips.
2. Description of the Related Art
Due to the increasing need for miniaturisation of portable devices such as MP3 players, mobile telephones and personal digital assistants, it has become increasingly important to implement both digital processing functions and related analog, especially audio, processing functions on the same chip or integrated circuit—a so called mixed signal chip. For this reason the implementation of analog functions on metal oxide semiconductor (MOS) based devices has become increasingly important.
A major problem with this technology however is that the digital and analog circuits require different characteristics from the same semiconductor technology. Digital circuits are fastest and consume least power and chip area when implemented using the shortest channel available in a given manufacturing technology. However this limits the supply voltage that can be applied to this digital circuitry without causing breakdown or premature wear-out of the small devices used. For example digital circuitry on a currently mature process might use a structure with 0.35 um drain-source spacing and 70 nm gate oxide thickness. However most analog circuits must operate in accordance with legacy standards, for example in order to provide a 2V rms signal for consumer standard audio Line Outputs, or possibly 5V rms for professional applications. Operation at these large signal swings is likely to continue to be necessary for some time in the future, to maintain the ratio of signal power to that of the thermal noise of the op amps and resistors in the signal path: a halving of the signal amplitude would require the noise power to be divided by four, reducing the required circuit impedances by a factor of four, and increasing the power required by the amplifiers, despite the lower supply voltage. Also reducing the signal level would increase the relative importance of extraneous noise and interference coupling into the circuitry. This is likely to require a rail to rail supply as high as 18V. To avoid large electric fields in the device, which cause breakdown or long-term reliability issues such as hot-carrier induced degradation of threshold voltage and transconductance, this requires a larger device structure, with typically a 3 um minimum drain-source spacing and a 350 nm oxide layer thickness.
Many widely available semiconductor manufacturing technologies today offer the possibility of selecting a thin or thick gate oxide thickness for selected transistors in each integrated circuit on the basis of additional photographic masking and processing steps. For instance one technology allows 70 nm gate oxides for 3.3V nominal (3.6V maximum) operation for core logic transistors, but 120 nm devices for 5V nominal (5.5V maximum) digital input and output devices. Such technology has been used for mixed signal circuits where the logic operates at 3.3V and all the analog circuitry operates at 5V. Similarly circuits such as LCD display drivers use 3.3V control logic and 18V nominal (19.8V maximum) (350 nm gate oxide) output stages. Allowing the logic to operate at lower voltage using smaller devices makes these devices smaller in chip area and hence cost and reduces the power consumed by the digital circuitry.
High performance audio op amps also require high open loop bandwidth so that the distortion inherent in their open-loop transfer characteristics may be suppressed across the audio band by negative feedback around the amplifier, even when this feedback is relatively light, to provide gain in the signal path. Suppression of distortion is important even at frequencies well above the human hearing limit of approximately 20 kHz. This is because distortion at these higher frequencies, for example up to 100 kHz, have an effect in the audible range (20 Hz-20 kHz). Also audio signals from delta-sigma digital-to-analog converters have quantisation noise components well above the audio band, which can intermodulate to produce audio band components unless the amplifier retains a linear closed-loop response to these high-frequency components. A wide closed-loop bandwidth is also necessary to avoid relative phase delays across the audio spectrum.
One of the major sources of noise in MOS technology is flicker noise. MOS devices such as transistors contain traps, due to impurities and inevitable imperfections in the crystal structure of the silicon, in or near the interface between the silicon and silicon oxide layers. The current in MOS devices typically travels substantially along this interface, and the traps charge and discharge randomly over time. This gives rise to a noise component of charge density at the oxide interface with an approximately 1/f power spectrum, i.e. with higher spectral density at lower frequencies.
For circuit analysis, this charge variance ΔQ may be regarded as an equivalent modulation of the gate voltage ΔVG where ΔVG=ΔQ/Cox, Cox being the capacitance from gate to channel across the gate oxide. Cox is inversely proportional to gate oxide thickness, so for the same charge variance, the equivalent gate voltage noise is proportional to gate oxide thickness. In practice, this is found to be the case, i.e. similar processes with different gate oxide thicknesses give gate flicker noise voltages increasing with gate oxide thickness.
It is also found that the flicker noise voltage is inversely proportional to the square root of the area of a MOS transistor. So one approach to reducing this noise is to increase the surface area, i.e. to increase the width and length of the transistor. However to improve flicker noise by say 6 dB would require four times the transistor area: significant further improvement in flicker noise rapidly leads to impractically large devices, both in terms of extra parasitic capacitances and in the chip area consumed and hence in cost of manufacture.
The contribution of flicker noise of a given transistor to the input-referred noise voltage of an amplifier can also be reduced by altering the gain from the transistor referred to the input, by altering its aspect ratio or altering its bias current. But this makes the design deviate from what would otherwise be considered the optimum in terms of the desired combination of area, power, and performance, and in practice there is again only a small improvement practically achievable without unduly compromising other design objectives.
Also chopper-stabilisation techniques could be incorporated to move flicker noise away to higher frequencies, where the noise can either be ignored or filtered out. However this adds to the complexity of the circuit, generally requiring the addition of multiple switches and clock generation and distribution circuitry, and tends to give spurious output signals at the chopping frequency and its harmonics.
So in general, for a given circuit topology, circuit specification, and manufacturing technology, there is a practical and economical lower limit to the flicker noise achievable.
A known circuit common in high performance audio amplifier applications is the differential folded cascode op amp circuit, a schematic for which is shown in FIG. 1a. This circuit offers low distortion, high gain, and wide bandwidth, which are desirable for hi fidelity sound reproduction. The operation of such circuits is well known to those skilled in the art, however the cascode arrangement essentially utilises a gain transistor (MP1 or MP2) together with a cascode transistor (MNC1 or MNC2) which effectively reduces the variation in voltage across its associated gain transistor (MP1 or MP2) in order for this to amplify changes in its input voltage in a linear fashion; thus reducing distortion. This topology also offers high voltage gain to the output lout and wide voltage compliance at this node, either for directly driving an output or to act as the input of a further op amp gain stage.
FIG. 1a shows a differential folded cascode amplifier structure using two cascode transistors (MNC1, MNC2) and constant current bias devices MNM1, MNM2. Since bias device MNM2 passes a constant current, all signal current from input device MP2 passes through cascode device MNC2 to the output lout. Similarly, signal current from MP1 passes through cascode device MNC1 rather than bias device MNM1, and is then mirrored by mirror devices MPB1, MPB2 to the output lout. Cascode devices MPC1 and MPC2 are inserted in series with the drains of MPB1, MPB2 to improve the output impedance and accuracy of this current mirror. Suitable bias voltages VCP1, VCN1, VBN1 are derived by other circuitry using standard techniques.
The folded cascode structure of FIG. 1b is a variation of this differential folded cascode amplifier. In this case previous bias devices MNM1 and MNM2 are reconnected as mirror devices with MNM1 being drain-gate connected, and cascode device MNC1 is also drain-gate connected, and previous mirror devices MPB1 and MPB2 now operate as constant bias current sources supplied with a suitable bias voltage VBP1. As before, signal current from MP2 flows through cascode device MNC2 to the output. However signal current from MP1 can no longer flow through cascode device MNC1, since this is now forced to operate at the constant current supplied by MPB1, so this signal current now flows though mirror device MNM1, where it is mirrored by MNM2 and thence flows through MNC2 to the output.
Whilst in these structures the flicker noise contribution of the cascode transistors MNC1, MNC2, MPC1 and MPC2 is small, in practical implementations of the circuits of FIG. 1a or FIG. 1b, it is found that the flicker noise contributed by MNM1 and MNM2 is one of the dominant components of audio frequency noise, with other flicker noise contributed by input devices MP1 and MP2 and by MPB1 and MPB2. This is particularly the case for high-voltage (say 18V) circuits where the amplifier is implemented with appropriately thick gate oxide (say 350 nm) MOS devices. As discussed above, the designer soon reaches a practical lower limit for this flicker noise. Yet there is an increasing requirement for lower and lower noise audio circuitry with better and better signal-to-noise ratio, i.e. lower noise and higher signal swings.